+
Trends in Disability in a Super-Aging Society:
Adapting the Future Elderly Model to Japan Jay Bhattacharya
with Brian Chen, Hawre Jalal, Karen Eggleston, Michael Hurley, and Lena Schoemaker
+ Preliminary
Please do not cite at this stage.
Comments welcome to Jay Bhattacharya
+ Background: Aging in Japan
Pronounced population aging
Low fertility and longevity decline to 87 million by 2060
40% of population over 65 by 2060
Old age dependency ratio – 70% by 2050 in Japan vs. 39% in the US
Highest proportion of elderly adults in the world
+ Policy Uses for A Future Eldery Model
Policymakers require projections of future need for
Long-term care and medical insurance
And other far-reaching areas: fertility;
immigration; economic policy and national security
+ Need to Account for Health Status of the Population
Few projections of Japan’s future elderly population
Re: disability, health, and need for long-term care
Most projections forecast age and sex ratios
Competing risks problem
+ Research Goal
To develop a demographic and economic simulation model to analyze impact of
Demographic change
Aging
Population Health
On
(Health spending)
Disability
Need for long-term care
The idea is to develop a way to explore the
implications of the best available data
+ Microsimulation Tracks Simulated Individuals Over Time
Survivors Etc.
New 51 year-olds
in 2016
Decease d
Survivors
Deceased Survivors
2016 costs Health &
functional status 2016 Health &
functional status, 2015
New 51 year-olds
in 2015
210,000 Older Japanese (age 51+)
in 2014
2015 costs 2014 costs
Deceased
+ Overall strategy
Estimate disease transition probabilities
Estimate mortality rates conditional on disease conditions
Construct a Markov microsimulation model based on the Future Elderly Model (FEM)
Project/simulate future medical conditions
and functional status/need for care (ADLs,
IADLs, cognition and social status measures)
+ Japanese Study of Aging and Retirement (JSTAR)
First longitudinal dataset on middle-aged and elderly Japanese
Two waves of interviews in 2007 and 2009
3,862 respondents between 47 and 75 in five Japanese cities
1,400 questions mirroring other surveys
conducted internationally (such as the HRS)
Self- reported information on physical or
mental limitations
+ Methods: Future Elderly Model
Health Transition Model
Logistic regression – to estimate probability of
transitioning into 19 mutually exclusive health states from 2007 to 2009
Focus on diseases most relevant and costly in a Japanese population
Treat all diseases as absorbing states
Disability Model
Ordered logistic regression to estimate ADLs/IADLs
Outcomes defined as having difficulty in 0, 1, 2, or 3+
(instrumental) activities of daily living
+ Methods: Health Transition Models
Health status measures
Heart disease, hypertension, hyperlipidemia, cerebrovascular disease, diabetes, chronic obstructive pulmonary disease, asthma, liver disease, ulcer, joint disease, bone
fractures/broken hip, osteoporosis, eye disease, bladder disease, mental health disorder,
dementia, skin disease, cancer and all other diseases
Other covariates
Age (demeaned), age2 demeaned, gender, smoking, obesity
+ Sample Selection
≥ 45 years
Non-missing variables in all outcomes and covariates
2,526 individuals for 2007, 2,659 for 2009,
and 1,854 individuals and 3,708 interview
years
+ Summary Statistics
Variable Obs Mean Std. Dev. Min Max
age 3,741 63.46 7.07 47.00 77.00
smoker 1,348 0.59 0.49 0 1
male if smoker 0.90
bmi 3,667 23.22 3.15 13.78 64.92
% bmi>=23.5 0.43
bmi if
overweight 1,577 25.94 2.55 23.51 64.92
female 3,744 0.50
+ Prevalence of Health Conditions
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 heart disease
hypertension hyperlipidemia CVD diabetes COPD asthma liver ulcer joint broken hip osteoporosis eye disease bladder mental health dementia skin cancer other
Prevalence
+ Disability Model
ADLs follow HRS definition
whether respondents are able to dress themselves, walk
around in their room, bathe, eat, get in and out of bed, and use Western-style toilets
IADLs
whether respondents are able to take public transportation alone, shop for daily necessities, prepare daily meals, pay bills, handle their own banking, make telephone calls and take medications
Social Engagement
visiting friends, being called on for advice, visiting sick
friends, and initiating conversations with younger individuals
Intellectual Engagement
filling out pension forms, reading the newspaper, reading books or magazines, and taking interest in the news
Care Receiving
Any help, help with physical care, help with chores
+ Summary Statistics – - Disability Model
ADL Prevalence
0 95%
1 3%
2 1%
3+ 2%
IADL Prevalence
0 93%
1 4%
2 1%
3+ 2%
Receives Aid Prevalence
Home Care 6%
Nursing Care 6%
Physical Help 22%
Help for Chores 43%
+ Mortality Data
Model requires estimates of mortality rates conditional on health status, age, and sex
Vital Statistics (Japanese MLHW)
Age/sex specific causes of death
But the available data provides a single cause of death, not the health status vector at death
+ Replenishing the Model with New 50 year olds
We draw new 50-year olds for future years by assuming that future 50-year olds will
have the same health distribution as current 50-year olds
Incoming 50-year olds are drawn from JSTAR
This assumption can be relaxed (as it has in
the US version of the FEM)
+ Cause-Specific Mortality Model
Steps to estimate conditional mortality model:
Specify rankings of diseases based on priority in death certificate reporting
Clinical ranking
Maximum likelihood model
Specify mortality rate as a linear function of disease prevalence
Specify loss function to estimate disease weights in conditional mortality function
Estimate disease weights using MHLW mortality
Separately by age groups and sex
+ Japan FEM Microsimulations
Once the model parameters are estimated, we simulate
Health Status
Physical/mental functioning
Of Japan’s 50+ elderly into the future (2010- 2040)
We are planning to extend the model to
other outcomes (health expenditures, etc.)
+ Model Checks
Simulate age structure of population for future years and check against official projections.
Backcasting exercise
Start model in 2000 and backcast for years 2001- 2010
Check population, mortality, and health predictions against actual data
This exercise is still in process
+
Results
+ Simulating Japan’s Future Aging
+ Simulating Japan’s Future
Population (vs. Official Projections)
+ Simulating Future Disease
Prevalence
+ Simulating Average Population
Prevalence of ADLs/IADLs/Others
+
Simulating Average Population Prevalence of Three or More Disabilities Purely Due to Changing Cohort Health, Holding the Age Distribution Constant at the 2010 Age Distribution+
The Impact of Obesity and Smoking on Future Disability:Simulating the Prevalance of ADL>=3 If All Japanese Quit
Smoking and No One Were Obese, or If Everyone Smoked and Was Obese
+ Limitations
Data censored at age 75-80
Leads to poor population forecasts for the oldest old.
Medical expenditures currently not
available
+ Next Steps
Additional data (NUJLSA)
Highest security version of JSTAR with links to claims (expenditures) data
Backcasting exercise
More sophisticated model for incoming 50
year olds
+ Summary
Population aging will lead to a much higher prevalence of chronic disease and
disability in the coming years
Even in the absence of population aging, chronic disease rates would rise in the coming decades
Prevention measures would offset the trend only slightly
Implications for policy
US FEM has already been presented to the US government and helps guide planning
We hope refined Japan FEM will also inform Japanese policy and planning